PCjr NMI and 8253 Timer interactions

[Chuck Guzis]

On 30 Sep 2009 at 11:31, CSquared wrote:

> Many years ago I tried a similar trick.  I think I wanted a 10
> millisecond interrupt rate instead of 55ms for an embedded DOS
> application, and I got it to work - sort of.  However, I soon
> discovered that many other things including the Borland delay()
> function depend on the 55ms interval, so I rather quickly abandoned
> that line of thinking.  I rather suspect your PCjr keyboard driver is
> making similar assumptions regarding the tick rate.

Several programs made this work on the PC, as long as the interval 
could be a sub-multiple of 55 msec.  So, for instance, if you could 
live with 11 msec. or 5.5 msec., this is how you would do it.

Hook the interrupt 9 vector and create your own ISR for 4 of 5 timer 
interrupts, then direct the the 5th to the normal vector.  All that's 
required in your ISR is any register saving required for your own 
needs, and a routine to issue an EOI to the 8259.   Works like a 
charm--everything runs just as it should.

However, on Mike's Peanut, there is no 8259.  Everything goes through 
the NMI handler--a very bad decision on IBM's part, seeing as how the 
8259 was one of the least expensive Intel 8x support chips.  A 
rudimentary prioritized interrupt system could have been implemented 
with a priority encoder and some other glue.

Cheers,
Chuck